You have something that has never been done before. ChatGPT has, therefore, not been fed any information about what it is, how it works, or what aspects are both commercially important and novel. You can try to supply that in your prompt and you might get a reasonable framework that could be edited to create a (possibly deceptively) good looking document.
Stand back and think about the actual case when a provisional actually helps you.
It is true that they are not subject to formal examination, so why worry about quality? Let’s look at the situation where there is a different outcome from having a provisional filed and not having it at all.
You have a non-provisional pending and you get a claim rejected that cites a publication that came out after the provisional but before the non-provisional. Unless this happens there is essentially no value in having filed the provisional.
The other case is you have made a public disclosure during that interim period and don’t put it on the IDS of the non-provisional because you think the provisional 100% covers you. Then it will not come up until you are in court trying to defend the patent and the sufficiently of the disclosure will be judged because the other side will bring it up.
You respond by referring to disclosure in the provisional that you assert fully supports the rejected claim. This is the first time the examiner (or court) opens the provisional. If the support in the provisional is solid, the publication is disqualified as prior art and you win.
However, the only time it matters that you even filed, the disclosure will be scrutinized exactly as it would be in a non-provisional. It is never literally “examined” but any disclosure you rely on will be examined seriously to see if the document teaches how to make and use the subject of the rejected claim, on its own.
If you file a provisional created by yourself or ChatGPT (an amazing feat of technology!) and act like you still need to keep the invention out of the public then it can only help in the unlikely case that it needs to be relied on.
I see that you do not want to tell ChatGPT the secret sauce. That makes sense since there is no confidentiality agreement between you and the people who have access to the ChatGPT prompts or may accidentally get access due to regurgitation. If you type in all the information it takes to make and use the invention you have likely broken novelty. In the US there is a one year grace period but elsewhere breaking novelty before filing kills a patent. For US purposes you could intentionally disclose fully in your prompt to try to get a better outcome.
Imagine you do not tell a patent attorney the secret sauce, do you think they can do a good job on your application regardless of their score on the USPTO bar?